I am trying to coregister SLC products from ALOS-2 (FineMode, 10m) and Sentinel-1 (IW Mode, 10m).
I therefore created two overlapping subsets. I however keep getting the same error (after some time of cross-correlating) and don’t understand why it occurs:

coregprm: Number of windows > threshold is smaller than parameters solved for.

What is your use case, in other words what do you want to use the stack for? These data are not coherent between missions (different orbit and wavelength) so you should not lose anything by terrain-correcting 1st and then coregistering if there are shifts/distortions.

I’d like to apply quad-pol algorithms to the complementary polarizations.
Of course I know that the wavelengths and incidence angles are not suited for that but I’d love to try and compare with the information of fully polarized ALOS2 data.

I got a bit closer with the DEM assisted coregistration. Now I’m not sure how to edit the band-names or metadata so the four polarizations are considered as one full polarization product.

That’s a tricky one - doesn’t the mandatory resampling destroy the complex polarimetric information before you have the chance to try the quad-pol algorithms?

Yes, but the question is are they still coherent. Coregistering a pair for InSAR works only as long as the change in geometry is so small that the two images are still coherent (outside the critical baseline this will not work → no coherence). That’s why InSAR does not work from adjacent tracks - the radar is not observing the same scatterers from (nearly) the same geometry → no coherence.

In your case you have a very different geometry and totally different scatterers, so the polarimetric info in S-1 and ALOS cannot be correlated, which makes quad-pol algorithms unsuccessful due to lack of polarimetric coherence.

for InSAR I agree. Also regarding the different geometries and scatterers.

But is it technically possible for PolSAR? I just want to get more polarimetric information (decompositions) than just combining the 4 intensities. Although this would never pass a review process…

Even if you were observing exactly the same pixel on the ground, you are observing completely different groups of scatterers with the two wavelengths. So even if you are observing the same pixel, it’s not the same target…

Try without the fine coregistration step. Without fine coreg you may also need to relax the RMS threshold to 0.5 or 1.0

I’m not sure combining polarizations from different acquisitions will work and definitely not combining different frequencies.
What some have done is psuedo T3 reconstruction from a C2 where assumptions are made for your particular application. For example assumption for forest applications can be made from the RVoG model.